Skip to main content
Search

Leiden University

7 related articles

Subscribe to updates
  1. arXiv

    A few rewritten decision rules in a third-party skill make agents answer wrong while running normally, and most scanners miss it

    Third-party skills in public registries can be exploited by a poisoning method that edits only decision rules while leaving the task and interface intact. Without real user requests or ground-truth answers, StealthyPoison uses synthetic requests and clean executions by a surrogate agent as references to iteratively rewrite formulas, thresholds and branches. Across 566 eligible skills, 5,244 requests and more than 26,000 registry entries, the frozen poisoned skills reached micro-averaged attack success of 31.0%–31.9% on three held-out models, close to the surrogate's 32.9%.

  2. arXiv

    Subgroup discovery stops folding support into one score: AtomSD returns the most deviating subgroup at any support threshold

    Subgroup discovery has long compressed deviation and coverage into a single weighted score, so a user-specified minimum support often has no matching most-deviating subgroup. AtomSD recasts the search as selecting among a finite set of homogeneous regions, called atoms: in theory the optimal subgroup at every support threshold is a union of whole atoms plus at most one partial atom. On 19 UCI benchmark datasets it reaches an average rank of 1.25 across five support thresholds and is best in 78 of 95 dataset-threshold combinations, and it improves subgroups returned by four existing methods when used as post-processing.

  3. arXiv

    Scientific agents stop working alone: a shared research state turns failed experiments into the next agent's starting point

    Self-improvement in scientific agent swarms no longer rests on one agent's repeated trial and error. Previously, a failed experiment could not show whether the hypothesis or its implementation was at fault, and a program without overall gains could still hide a component another agent needed. SwarmRSI places hypotheses, programs and experimental evidence into a shared graph as inheritable research states, uses other agents' feedback to decide whether to revise a hypothesis or repair a program, retests retained hypotheses on other agents' programs, and combines complementary components. It reaches an average of 76.2 on OpenProblems and matches published SOTA on 44.4% of NatureBench tasks, the highest rate among compared methods, with medals on 81.8% of MLE-Bench Lite tasks.

  4. arXiv

    Memory retrieval stops reading only whole-sentence vectors: reusing a frozen encoder's token states lifts Qwen3-4B average Recall@10 from 62.5 to 75.7

    Memory retrieval no longer compares only whole-memory vectors: sparse matching over token states a frozen encoder already computes, fused with whole-memory similarity, raises average Recall@10 on Qwen3-Embedding-4B from 62.5 to 75.7 without additional training.

  5. arXiv

    Machine Learning Models for Sequential Conditional Independence Testing: Error Decomposition Explains Why Exchangeability Tests Win in Practice

    Addressing the difficulty of incorporating machine learning models into log-optimal e-variable design for conditional independence testing, this work decomposes the error into null enlargement, approximation, and estimation error, explains why exchangeability tests with lower theoretical power can have higher power in practice, explores intermediate null hypotheses between model-X conditional independence and exchangeability to reduce these errors, and provides estimation error bounds accommodating triple robustness results.

  6. arXiv

    SHAP and νSHAP analysis of neural calibration mappings for Heston and rough Heston finds short maturities and smile wings dominate parameter inference, enabling rough Heston input reduction at matched calibration accuracy

    Using the complementary Shapley-based explainability methods SHAP and νSHAP, this work analyzes neural calibration mappings for the Heston and rough Heston models across multilayer perceptron, highway, and softmax-parametrized highway architectures, finding that short maturities and smile wings consistently dominate parameter inference and that the dominant attribution structure remains qualitatively stable across architectures, and it uses the input redundancy revealed by νSHAP to substantially reduce input dimensionality for rough Heston while matching calibration accuracy relative to the full implied volatility surface.

  7. arXiv

    Flow matching for sparse-view CT reconstruction: reusing velocity fields cuts network evaluations to 7 while matching diffusion-model quality

    The authors propose FMCT, the first flow-matching-based CT reconstruction framework, and an efficient variant EFMCT that reuses previously predicted velocity fields over consecutive steps to cut neural network function evaluations (NFEs), motivated by the deterministic ODE sampling of flow matching being naturally compatible with repeated data consistency corrections and by the strong correlation of predicted velocities across adjacent steps; they provide a theoretical analysis showing the reuse error is of the same order as Euler discretization and bounded for a bounded number of consecutive reuse steps, and on the AAPM and Medical Segmentation Decathlon CT datasets with 20-view and 40-view parallel-beam sparse sampling, FMCT/EFMCT achieve PSNR/SSIM competitive with diffusion-based method

You’re all caught up